- cancel_on_shutdown
- robustification dans backend
- idle workers should shutdown while executor should restart missing worker on submit
- modification parameters ==> recreate pool
- debug lock on start executor
- support python2.7 (kwargs.pop ? ) regardé concurrent future package
- remove appveyor logs
- cutsomizable pickler for the queues in ProcessPoolExecutor
- cleanup compat files
- refactor thread to optimize communication (_feeder thread) #12
- Test for errors and crashes in job dispatch by the apply async callback
- Configure Appveyor to run the test under windows (see joblib one)
- Add travis to run the test under linux/ mac
- Unskip numpy freeze test once Spawn Processes are implemented
- Test the callbacks call with ProcessPoolExecutor
- Unit test semaphore and fd managment
- Make cython model only built in test that highlight the OpenMP freeze with multiprocessing fork
- See if Rpool pass Pool test suite from cpython
- Do not use the result queue to wakeup _queue_management
- Perfomance optimization for _queue_management_thread wakeup
- Add broken process pool detection thread (after #12 is solved)
- Create process with subprocess.Popen and os.pipe for posix
- Test processes with cpython test suite
- Pass test rpool with such processes
- [-] automatic benchmark script launched in CI after test (#11)
- benchmark large messages (string)
- Implement new backend based on loky
- Make new backend auto mmemap large numpy